Abstract: |
Abstract. We hypothesize that the relationship between parameter set-tings, specically parameters controlling mutation, and performance isnon-linear in genetic programs. Genetic programming environments havefew means for a priori determination of appropriate parameters values.The hypothesized nonlinear behavior of genetic programming createsdiculty in selecting parameter values for many problems. In this paperwe study three structure altering mutation techniques using parametricanalysis on a problem with scalable complexity. We nd through param-eter analysis that two of the three mutation types tested exhibit nonlin-ear behavior. Higher mutation rates cause a larger degree of nonlinearbehavior as measured by tness and computational ert. Characteriza-tion of the mutation techniques using parametric analysis conrms thenonlinear behavior. In addition, we propose an extension to the existingparameter setting taxonomy to include commonly used structure alteringmutation attributes. Finally we show that the proportion of mutationsapplied to internal nodes, instead of leaf nodes, has a signicant ecton performance. |
